PunGraph

A compact collection of English pun data and phonetic-similarity resources for pun detection, interpretation, generation, and retrieval experiments.

Pun-Inference-Public/
├── KG/
│   └── phonetic_similarity_table_replaced_cleaned.txt
├── WebDataset/
│   ├── final_heter.json
│   └── final_homo.json
└── README.md

final_homo.json contains a top-level data array. Each item has the following fields:

Example:

{
  "id": 1,
  "sentence": "A baseball player can sell himself to a new team if he has a good pitch.",
  "Pun": "pitch",
  "pun_word": "throw, sales proposal"
}

final_heter.json also contains a top-level data array.

Example:

{
  "id": 1,
  "sentence": "Make like a banana and spit",
  "Pun": "spit",
  "Real": "split"
}

phonetic_similarity_table_replaced_cleaned.txt contains one pair per line:

word_1|word_2|similarity_score

The score ranges from 0.2 to 1.0 in the current release, where a larger value indicates greater phonetic similarity.

a-bomb|bomb|0.75
a-bomb|h-bomb|0.8333333333333334

No package installation is required to inspect the data. Clone the repository and load the JSON files with Python's standard library:

git clone https://github.com/ysu132/Pun-Inference-Public.git
cd Pun-Inference-Public

import json
from pathlib import Path

root = Path("WebDataset")

with (root / "final_homo.json").open(encoding="utf-8") as file:
    homographic = json.load(file)["data"]

with (root / "final_heter.json").open(encoding="utf-8") as file:
    heterographic = json.load(file)["data"]

print(f"Homographic examples: {len(homographic):,}")
print(f"Heterographic examples: {len(heterographic):,}")
print(heterographic[0])

Read the similarity table as a stream so the entire file does not need to be loaded into memory:

from pathlib import Path

table = Path("KG/phonetic_similarity_table_replaced_cleaned.txt")

with table.open(encoding="utf-8") as file:
    for line in file:
        word_1, word_2, score = line.rstrip("\n").split("|")
        score = float(score)
        # Use the pair here.

• The data is in English.
• Field names are case-sensitive (Pun and Real).
• Record IDs are local to each JSON file; do not assume they are globally unique across the repository.
• Review and normalize the data as appropriate before using it in production or model evaluation.

If you use this repository in your work, please cite the repository URL. A formal paper citation will be added here when available.

No license file is currently included. Until a license is added, the repository does not grant permission to copy, modify, or redistribute its contents beyond what is allowed by applicable law. Please contact the repository owner for usage terms.